yobx.torch.patch_enum#

class yobx.torch.patch_enum.TransformersPatchEnum(*values)[source]#

Selects which transformers-related patches apply_patches_for_model() should apply. Members can be combined with |.

  • NONE — no transformers patches are applied.

  • YOBX_PATCH — applies yobx’s own transformers patches (rotary embedding wrappers, flattening helpers, …).

  • TRANSFORMERS_PATCH — toggles the ONNX-export switches that transformers itself ships (config.onnx_export, module.prepare_for_onnx_export_()).

  • ALL — shortcut for YOBX_PATCH | TRANSFORMERS_PATCH.

yobx.torch.patch_enum.coerce_transformers_patch(value: bool | TransformersPatchEnum | str) → TransformersPatchEnum[source]#

Backward-compat conversion: True -> ALL, False -> NONE.